Supplementary material to manuscript:
Looking upstream to prevent HIV transmission: can interventions with sex
workers alter the course of HIV epidemics in Africa as they did in Asia?
Richard Steen, Jan AC Hontelez, Andra Veraart, Richard G White, Sake J de Vlas
Summary of modeling approach
The underlying model structure, data and assumptions behind this work come from an
STDSIM quantification developed for Kisumu as part of the 4-cities study, which has
previously been described in detail. [1] This supplemental material expands on the
Methods section of the paper to document (1) changes made to the original quantification
for this study, (2) model fitting, and (3) explorations of alternative scenarios.
1. Changes to the original quantification
As described in Methods, four main improvements were made to the quantification of sex
work used in the 4-cities study. We updated HIV transmission probabilities used in 4-cities
modelling to lower values that resulted from a recent systematic review. [2] Data and
assumptions related to sex work were reviewed and updated, and the simulation of sex
work refined to permit more detailed estimates and analyses. This involved changing the
quantification of sex work, using the mean number of clients per week (1.6) from survey
data, rather than the median (1.0) as reported by Morison and used by Orroth. [3,1] We
also refined the simulation of commercial sex to include three categories of sex workers
with different rates of partner change (numbers of clients). Finally, STI treatment
assumptions were updated based on recent programme evaluation data. [4] After making
these changes, STI transmission probabilities were refitted in order to arrive at observed
STI prevalences. All other quantifications remain unchanged.
1.1. Updating and refining the quantification of sex work. We reviewed all original
parameter assumptions related to sex work in the original quantification for Kisumu and
made several changes. Four parameters determine the overall level of sex work – sex
worker and client population sizes and activity (number of clients for sex workers, number
of sex worker visits for clients) per unit time. These parameters are closely linked in
STDSIM with client demand driving sex worker supply. As in the original study, we
consider data on sex worker parameters (from mapping studies and behavioural surveys)
more reliable than for male clients (where there are important biases). We thus quantified
the client parameters to give sex worker population sizes and activity levels that
approximate values from the data.
The main change from the original study for sex workers was increasing their number of
clients per week from the median of 1.0 as reported by Morison to the mean value of 1.6
from 4-cities survey data. [1,3] More recent (2006) survey data from Kisumu using
respondent-driven sampling suggest even higher client numbers (median = 3 per week,
interquartile range = 2-5). [7,8] We then extended STDSIM to permit disaggregation of sex
workers into 3 activity groups. This was done in a way to maintain overall sex worker
population size and mean client numbers as already described. The assumptions are
examined further in 3.3.
One important parameter set – client activity – was increased in order to reach sex worker
population sizes and reported activity as informed by data (sex worker selection is random,
proportional to sex workers’ defined client numbers). Client activity includes two
components – client population size as a proportion of adult males and their visiting
behaviour or frequency of visiting sex workers, which varies by marital (regular partner)
status. Reliable data on both of these are sparse due to inherent biases of household
surveys. [5,6] Our approach was to maintain the client population sizes from the 4-cities
and to increase visiting behaviour (demand side) – from 1 to 3 visits per year for less
active, and from 12 to 18 times per year for more active clients – in order to support the
sex worker parameters (supply side).
The overall size of the sex worker population, an output variable, remained essentially the
same (2.2% of women 15-49 compared to 2.1% in original). This is still lower than data
(3.0%) suggest, however, which may underestimate the contribution of sex work to
disease transmission. [9] Increasing sex worker population size further would have
required raising client numbers and/or their rate of visiting sex workers even more than we
did. Taken together, these changes result in ‘more’ sex work than in the original 4-cities
quantification although this may still be underestimated considering the most reliable data
(sex worker reported client numbers and population size). The effect of varying these
assumptions is examined in 3.1 and 3.2.
1.2. Updating transmission probabilities and STI treatment assumptions. A recent
systematic review suggests that HIV transmission probabilities are lower than those used
in the original 4-cities modeling study. [2 Boily] We thus reduced HIV transmission
probabilities to values from the systematic review (see Table S1).
Table S1. HIV transmission probabilities used in the model (from Boily et al.) [2]
HIV
primary
HIV
asymptomatic
HIV
symptomatic
AIDS
M –> F
1.426%
0.095%
0.285%
0.713%
F –> M
0.713%
0.047%
0.143%
0.356%
STI cofactor estimates for HIV transmission and acquisition were not changed. These
remain at 3x for gonorrhea and chlamydia, 7.5x for ulcerative syphilis, 25x for ulcerative
chancroid and primary HSV-2, and 10x for non-primary HSV-2 ulcers.
Two STI treatment interventions in the original quantification – STI syndrome case
management introduced in 1992 and an STI screening intervention introduced for sex
workers in 1993 – were removed. These changes were based on more recent data, which
describe improvements in the 1990s to be marginal and short-lived, virtually collapsing in
2001 following the end of World Bank project funding which supported STI drug provision.
[4]
STI transmission probabilities were then refitted within a priori ranges (based on estimates
with wide confidence intervals) maintaining the 2:1 ratio for male-to-female compared to
female-to-male transmission, as was done in the original 4-cities model. These crude fits
resulted from 10% steps of deviation from the original quantifications in such a way that (1)
the model predictions fitted the prevalences for M, F and CSW best, and (2) transmission
probabilities stayed within the fitting range set for the 4-cities study. These changes are
summarised in Table S2.
Table S2. Original 4-cities and revised STI transmission probabilities and refitted STI outputs
Transmission probabilities
Syphilis
(average over first 6
months)
Gonorrhoea
(14 weeks)
Chlamydia
(14 weeks men, 52
weeks women)
Chancroid
(11 weeks)
M–>F revised
0.0875
0.156
0.126
0.18
F–>M revised
0.0438
0.078
0.063
0.10*
(% of original)
50%
60%
50%
80%
STI prevalences 1997 (data and model outputs)
Syphilis
Male
Female
Sex
worker
Gonorrhoea
Chlamydia
Chancroid
Data (CI)
3.1 (1.5)
0 (1.2)
2.6 (1.4)
no data
Original
7.7
1.4
2.8
0.0
Revised
5.0
0.8
1.6
2.0
Data (CI)
3.9 (1.3)
0.9 (0.6)
4.5 (1.4)
no data
Original
8.7
2.1
6.0
0.0
Revised
5.2
1.0
3.5
1.2
Data
11.0
13.0
8.0
no data
Original
20.6
8.8
11.9
0.0
Revised
15.5
6.9
13.9
11.8**
* Lower limit of a priori range
**Note that while there are no data on chancroid, indirect data supports its presence. Morison reports very
high prevalence of genital ulcers (29%) among Kisumu sex workers based on clinical examination.
2. Fit of the model to HIV prevalence data
The changes described above influenced the fitting of our revised quantification, which has
a somewhat lower prevalence trend than the original, but a good fit to the original study
data (Figure S1).
Fig S1. Original 4-cities and revised quantification for Kisumu
3. Alternative scenarios
3.1 Effect of varying key parameter values
Several parameter assumptions were subject to further exploration to assess whether
changes in their values would affect the main conclusions of the study. The primary
outcome of interest is HIV prevalence in 2010, ten years after the time when interventions
are introduced in the paper. For example, in the paper’s quantification, HIV prevalence is
estimated to be 22.8% in 2010 with no intervention. After removing transmission from the
40% of most active sex workers (HM) in 2000, this declines to 16.4%. Table S3
summarizes effects of varying selected parameter values on these outcomes.
Table S3. Effect of varying select parameter values on HIV prevalence and intervention outcomes
Parameter
Parameter
value
Reference
HIV
prevalence
2010
% change after
blocking
transmission in
sex work (HM)
22.8%
28.2%
Conclusions
1. Sex worker and client
population sizes.
Sex worker population size may
be underestimated in the
quantification, suggesting that
client demand is also low.
Decrease by
2/3
14.9%
24.4%
Varying assumptions about
population size affects overall
HIV transmission level and
proportional effect of
interventions, but not main
conclusions.
Increase by
1.5
31.4%
33.2%
2. Sex worker partner change
(mean number of clients in last
week), which may be
underestimated in the
quantification. [4,5]
Reference (mean=1.6)
Increase by
1.5
27.6%
33.7%
Increasing sex worker client
numbers increases the
contribution of sex work to
HIV transmission and the
effect of interventions,
strengthening conclusions.
3. Sex worker partner change
(mean client number by sex
worker subgroup, skewness of
distribution).
Reference parameter values were
taken from distribution of sex
worker survey responses (mean =
1.6 clients per week as in #2).
Reference (L=0.55, M=2.19,
H=5.67 )
Decrease
(L=1, M=2,
H=4)
18.2%
19.7%
Varying assumptions about
the skewness of the sex
worker partner change
parameter affects HIV
transmission level and
proportional effect of
interventions, but not main
conclusions.
4. Condom use in sex work at
baseline.
This parameter value is based on
reported condom use at baseline,
which may be overestimated due
to social desirability bias.
Decrease by
2/3
27.8%
35.7%
At lower baseline rates of
condom use, the relative
effect of interventions on
transmission is greater.
5. STI transmission probabilities
and treatment interventions.
These were updated from original
4 cities parameter values as
described in section 1.
Original STI
transmission
probabilities;
same
treatment
assumptions
until 2000.
26.4%
28.8%
Varying STI transmission
probabilities and treatment
increases STI and HIV
transmission levels but does
not alter main conclusions.
6. At a specific level of overall
client demand (see #1 above), are
assumptions about the number of
sex worker visits by clients in two
risk categories (H/L). Original
quantification uses H=12, L=1,
and revised uses H=18, L=3.
H= 24/year,
L=2/year
24.3%
31.7%
H=12/year,
L=4/year
19.2%
22.9%
Increasing the difference
between high and low-activity
clients increases HIV
transmission but does not
alter main conclusions.
3.2 Effect of sex worker interventions in context of ART rollout
In addition, as described in the main paper, we ran several simulations in a quantification
that included progressive roll-out of ART as has been reported for Kenya. [10]
HIV and ART in STDSIM
Based on data from Orange Farm, South Africa, we assumed the initial (HIV-negative)
CD4 cell count in the population to follow a lognormal distribution with median 7.02
(equivalent to 1116 cells/µL) [11]. CD4 cell counts to decline by 25% during acute
infection, and decline linearly over the other stages until the CD4 cell count reaches 0.5%
of its initial value, after which an individual dies of AIDS. Therefore, based on the average
HIV-negative CD4 cell counts and duration of HIV infection, an individual will on average
reach a CD4 cell count of ≤350 cells/µL after about 6 years following infection, and at that
point will have a remaining life expectancy of about 4 years.
In the model, ART coverage is the result of two components: (i) an individual's demand for
ART as a function of HIV-disease stage, and (ii) the capacity of the health system to meet
this demand. ART coverage in our model is the ART demand met by the capacity of the
health system. We assumed the ART demand function to be the same as previously
estimated for the Hlabisa sub district of KwaZulu-Natal, South Africa [12], in which about
30% of all infected individuals start to seek care for HIV well after their CD4 cell counts
drop below 200 cells/µL. We fitted the model predictions to observed ART coverage levels
over the period 2004 -2010 by performing a grid-search on three parameter: start-year of
ART scale-up, rate of ART scale-up, and ART scale-up function (3 options: linear, squareroot, or quadratic). We optimized predicted ART coverage by calculating the Mean
Squared Error of predicted ART coverage in the model compared to coverage data
reported by WHO [13]. We assumed eligibility criteria to change from ART at ≤200 cells/µL
to ≤350 cells/µL in mid-2011, and the ART scale-up to continue according to the estimated
scale-up pattern for the years 2012 - 2050 in the baseline.
For Kenya, the optimal fit was given by a start of the ART scale-up in 2003, and by 2010,
50% of those eligible (at CD4 cell counts of ≤350 cells/µL) are on treatment. Universal
access (coverage of >80%) in the model is achieved in 2014 and maintained thereafter.
Figure 5 in the manuscript shows the more rapid decline in HIV prevalence when condom
use among sex workers (HM) increased in 2000 as compared to the ART only scenario.
We also ran the same simulations with later introduction of sex worker condom use
interventions. Figure S2 shows HIV prevalence estimates in the ART scenario where sex
worker condom use increases to 70% in 2012.
We conclude that the main findings of our study are also valid in settings with ART scaleup, and that the prevention effects of sex worker interventions and ART appear to be
largely independent and complementary.
Fig S2. ART roll-out with sex worker condom use intervention (70% HM) delayed until 2012
References
1. Orroth KK, Freeman EE, Bakker R, et al. Understanding the differences between contrasting
HIV epidemics in east and west Africa: results from a simulation model of the Four Cities
Study. Sex Transm Infect 2007; 83 Suppl 1:i5-16
2. Boily MC, Baggaley RF, Wang L, Masse B, White RG, Hayes RJ, et al. Heterosexual risk of
HIV-1 infection per sexual act: systematic review and meta-analysis of observational studies.
Lancet Infect Dis 2009; 9(2):118-29
3. Morison, L., Weiss, H. A., Buvé, A. et al. Commercial sex and the spread of HIV in four cities in
sub-Saharan Africa. AIDS 2001; 15 Suppl 4:S61–9
4. Cheluget B, Joesoef MR, Marum LH et al. Changing Patterns in Sexually Transmitted Disease
Syndromes in Kenya After the Introduction of a Syndromic Management Program. Sex Transm
Dis 2004; 31:522–525
5. Korenromp EL, Van Vliet C, Bakker R, De Vlas SJ, Habbema JDF. HIV spread and partnership
reduction for different patterns of sexual behaviour: a study with the microsimulation model
STDSIM. Math Popul Studies 2000; 8:135–173
6. Buvé, A., Caraël, M., Hayes, R. J. et al. Multicentre study on factors determining differences in
rate of spread of HIV in sub-Saharan Africa: methods and prevalence of HIV infection. AIDS
2001; 15 Suppl 4: S5–14
7. Vandenhoudt H, Langat L, Zeh C, et al. Community based study of HIV, STI, and sexual
behavior among female sex workers in Kisumu, Nyanza Province, Kenya. 18th ISSTDR 2009,
London, 28 June–1 July
8. Langat L, Vandenhoudt H, Menten J, et al. Estimating the size of the female sex worker
population in Kisumu, Western Kenya. 18th ISSTDR 2009, London, 28 June–1 July
9. Vandepitte, J., Lyerla, R., Dallabetta, G. et al. (2006). Estimates of the number of female sex
workers in different regions of the world. Sex. Transm. Infect., 82(Suppl. 3), iii18–25
10. Hontelez JA, de Vlas SJ, Baltussen R, et al. The impact of antiretroviral treatment on the age
composition of the HIV epidemic in sub-Saharan Africa. AIDS 2012; 26 Suppl 1: S19-30
11. Williams BG, Korenromp EL, Gouws E, Schmid GP, Auvert B, Dye C. HIV Infection,
Antiretroviral Therapy, and CD4+ Cell Count Distributions in African Populations. J Infect Dis
2006; 194 (10): 1450-1458 doi:10.1086/508206
12. Hontelez JA, de Vlas SJ, Tanser F, et al. The impact of the new WHO antiretroviral treatment
guidelines on HIV epidemic dynamics and cost in South Africa. PLoS One 2011; 6(7):e21919
13. Hontelez JA, Lurie MN, Bärnighausen T, et al. Elimination of HIV in South Africa through
expanded access to antiretroviral therapy: a model comparison study. PLoS Med 2013; 10(10):
e1001534. doi:10.1371/journal.pmed.1001534
14. World Health Organization (2010) Towards universal access: scaling up priority HIV/AIDS
interventions in the health sector: progress report 2010. Geneva: World Health Organization